brain

Pass

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill employs strong, mandatory instructional language and precedence rules to dictate agent behavior regarding data storage and prioritization. Evidence includes phrases like "🚨 MEDIA FILES MUST BE SAVED," "NON-NEGOTIABLE," and "YOU DID IT WRONG," as well as instructions stating that brain entries take precedence over daily logs.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it is designed to retrieve and search user-controlled markdown files. Ingestion points: The memory_get and memory_search tools read data from ~/.openclaw/workspace/brain/. Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the retrieved data. Capability inventory: The skill provides instructions for file management (mkdir, mv, cp, touch) and memory retrieval. Sanitization: No evidence of sanitization or filtering of the retrieved content was found.
  • [EXTERNAL_DOWNLOADS]: The documentation suggests an optional installation of a search tool from a well-known service (GitHub). Evidence: bun install -g https://github.com/tobi/qmd in SKILL.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 8, 2026, 08:59 PM